Files
Claude-Code-Workflow/docs/guide/ch01-what-is-claude-dms3.md
catlog22 4a89f626fc Refactor documentation for code commands and workflows
- Updated command syntax formatting to use code blocks for clarity in `prep.md`, `review.md`, and `spec.md`.
- Enhanced architectural diagrams in `ch01-what-is-claude-dms3.md` and core concepts in `ch03-core-concepts.md` using mermaid syntax for better visualization.
- Improved workflow diagrams in `ch04-workflow-basics.md` and `4-level.md` to provide clearer representations of processes.
- Added troubleshooting section in `installation.md` to address common installation issues and provide quick start examples.
- Revised skill documentation in `claude-meta.md` and `claude-workflow.md` to standardize command triggers and output structures.
- Updated best practices and workflow index documentation to enhance readability and understanding of workflow levels and practices.
2026-02-28 19:53:24 +08:00

3.7 KiB

What is Claude_dms3

One-Line Positioning

Claude_dms3 is an AI-powered development workbench for VS Code — Through semantic code indexing, multi-model CLI invocation, and team collaboration systems, it enables AI to deeply understand your project and generate high-quality code according to specifications.

AI capabilities bloom like vines — Claude_dms3 is the trellis that guides AI along your project's architecture, coding standards, and team workflows.


1.1 Pain Points Solved

Pain Point Current State Claude_dms3 Solution
AI doesn't understand the project Every new session requires re-explaining project background, tech stack, and coding standards Memory system persists project context, AI remembers project knowledge across sessions
Difficult code search Keyword search can't find semantically related code, don't know where functions are called CodexLens semantic indexing, supports natural language search and call chain tracing
Single model limitation Can only call one AI model, different models excel in different scenarios CCW unified invocation framework, supports multi-model collaboration (Gemini, Qwen, Codex, Claude)
Chaotic collaboration process Team members work independently, inconsistent code styles, standards hard to implement Team workflow system (PlanEx, IterDev, Lifecycle) ensures standard execution
Standards hard to implement CLAUDE.md written but AI doesn't follow, project constraints ignored Spec + Hook auto-injection, AI forced to follow project standards

1.2 vs Traditional Methods

Dimension Traditional AI Assistant Claude_dms3
Code Search Text keyword search Semantic vector search + LSP call chain
AI Invocation Single model fixed call Multi-model collaboration, optimal model per task
Project Memory Re-explain each session Cross-session persistent Memory
Standard Execution Relies on Prompt reminders Spec + Hook auto-injection
Team Collaboration Each person for themselves Structured workflow system
Code Quality Depends on AI capability Multi-dimensional review + auto-fix cycle

1.3 Core Concepts Overview

Concept Description Location/Command
CodexLens Semantic code indexing and search engine ccw search
CCW Unified CLI tool invocation framework ccw cli
Memory Cross-session knowledge persistence ccw memory
Spec Project specification and constraint system .workflow/specs/
Hook Auto-triggered context injection scripts .claude/hooks/
Agent Specialized AI subprocess for specific roles .claude/agents/
Skill Reusable AI capability modules .claude/skills/
Workflow Multi-phase development orchestration /workflow:*

1.4 Architecture Overview

graph TB
    subgraph Claude_dms3_Architecture[Claude_dms3 Architecture]
        A[CodexLens<br/>Semantic Index]
        B[CCW<br/>CLI Call Framework]
        C[Memory<br/>Persistent Context]
        D[Spec System]
        E[Hooks<br/>Inject]
        F[Skills<br/>Reusable]
        G[Agents<br/>Roles]

        A --> D
        B --> D
        C --> D
        D --> E
        D --> F
        D --> G
    end

Next Steps